基于Parzen窗算法的图像视觉显著目标识别算法  

Visual Salient Object Recognition Algorithm Based on Parzen Window Algorithm

作　　者：董薇[1] 窦立君[1] DONG Wei;DOU Li-jun(Nanjing Forestry University,Jiangsu Nanjing 210037,China)

机构地区：[1]南京林业大学,江苏南京210037

出　　处：《计算机仿真》2024年第5期214-219,共6页Computer Simulation

基　　金：国家自然基金(61871444)。

摘　　要：图像的复杂度与日俱增导致图像语义信息难以自动化获取,传统目标显著性检测方法获取图像信息存在稳定度低,准确度差的问题,为此提出一种基于改进Parzen窗目标位置估计优化算法,通过对估计区域进行双特征提取与融合,构建出PAR-SVM图像显著目标分类识别模型。模型首先对图像进行二值处理与形态学处理,并利用改进Parzen窗算法对显著性目标进行密度位置估计;然后提取图像中显著目标位置的G、H特征,并进行有机融合后规划数据集;最后基于数据驱动的方法,构建出PAR-SVM图像显著目标识别模型,并使用交叉验证对模型参数优化。实验一消融仿真结果表明:通过优化策略的叠加有效的提高了模型的准确率,与未优化前相比提升了19.12%。实验二对比仿真结果表明:与其它5类分类识别算法相比,在SOD数据集上,PAR-SVM算法的准确率高达86.5%,平均提高了3.14%,稳定性高达86.0%,平均提高了2.3%。综上所述,基于改进Parzen窗算法的图像显著目标识别模型在提高检测准确率的同时,也提高了模型的稳定性能。It is difficult to automatically acquire image semantic information due to the increasing complexity of the image,and the traditional target saliency detection method for acquiring the image information has the problems of low stability and poor accuracy,so the invention provides an optimization algorithm based on improved Parzen window target position estimation,and a PAR-SVM model for classification and recognition of salient objects in images is constructed.Firstly,the image was processed by binary value processing and morphology,and the density position estimation of the salient target was carried out by using the improved Parzen window algorithm,then the G and H features of the salient target position in the image were extracted,and the data set was planned after organic fusion,and finally the salient target recognition model of PAR-SVM image was constructed based on the data-driven method,and the model parameters were optimized by cross-validation.The simulation results of the first ablation experiment show that the accuracy of the model is improved by 19.12%compared with that before optimization.The simulation results of the second experiment show that the accuracy of the PAR-SVM algorithm is as high as 86.5%on the SOD data set,with an average increase of 3.14%,and the stability is as high as 86.0%on the SOD data set,with an average increase of 2.3%,compared with other five classification and recognition algorithms.To sum up,the image salient object recognition model based on the improved Parzen window algorithm in this paper improves the detection accuracy as well as the stability of the model.

关 键 词：目标识别 图像处理 显著性检测 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]